AMD's $243 Billion AI Disaster...What Happened?

Summary

Highlights

• Despite dominating CPUs, AMD struggled with GPUs, particularly against NVIDIA's AI-centric strategies 🎯.
• NVIDIA's CUDA software created a powerful ecosystem for AI development, attracting numerous researchers ⛓️.
• AMD's late AI investments, like acquiring Xilinx, failed to keep pace with NVIDIA's advancements 🐢.
• Financial losses and missed expectations plagued AMD despite bullish investments in AI 🔍.
• Lisa Su's leadership was crucial in diversifying AMD's product line, yet AI offerings lagged ⚙️.

Key Takeaways

• AMD stumbled in the AI race, losing ground to NVIDIA's superior GPU and software ecosystem 🎮.
• Legal battles with Intel over anti-competitive practices in the 2000s distracted AMD from AI developments 🤯.
• Lisa Su's leadership revitalized AMD, winning back market share with innovations like Ryzen 💪.
• Investments in AI like the acquisition of Xilinx cost AMD heavily, yet failed to yield expected returns 💸.
• NVIDIA's early focus on AI with CUDA established an insurmountable lead in the research community 🏅.

Overview

Chapters

• 00:00 - 01:00: Introduction and AMD's Decline The chapter discusses the impact of the AI boom on the semiconductor industry, particularly focusing on NVIDIA's success in becoming one of the most valuable companies due to its GPU designs. In contrast, AMD, which also designs GPUs, has not seen the same level of success and has faced a decline in its share value from early 2024 to early 2025.
• 01:00 - 03:00: AMD vs. Intel in the 2000s The chapter titled 'AMD vs. Intel in the 2000s' begins by discussing the significant market movements involving a GPU company during a period of AI hype, which saw a drastic reduction in market capitalization by almost $200 billion. It raises the question of what's happening with AMD and sets the stage to answer this by reflecting on the historical competition between AMD and Intel. In the 2000s, both AMD and Intel were heavily engaged in competition within the CPU market, competing to produce the central processing units that power computers and devices.
• 03:00 - 04:30: AMD's Acquisition of ATI and Financial Struggles The chapter discusses the competition between AMD and Intel in the CPU market. Intel, having its own factories, was able to produce processors at a lower cost and higher volume than AMD, which relied on GlobalFoundries. Additionally, Intel had strong partnerships with major customers like Dell and HP. Moreover, there were allegations that Intel was paying billions to these customers to avoid using AMD's processors, leveraging its larger financial resources to hinder AMD's competitiveness.
• 04:30 - 07:00: Lisa Su's Leadership and Diversification Strategy The chapter discusses AMD's legal actions against Intel, accusing them of engaging in anti-competitive practices. The Japan Fair Trade Commission discovered that Intel was providing rebates to five major Japanese computer manufacturers, though these companies were not named in the public order.
• 07:00 - 09:00: AMD's Comeback with Ryzen The chapter titled 'AMD's Comeback with Ryzen' delves into the legal and competitive landscape faced by AMD in its struggle against Intel. It highlights AMD's lawsuit against Intel for allegedly violating Japanese competition law. AMD contends that Intel has been utilizing 'first-dollar rebates'—a practice where manufacturers receive retroactive discounts, if they meet Intel's exclusivity demands at the end of a quarter, essentially pressuring the manufacturers to exclusively use Intel processors. The narrative also touches upon the evolving competition between AMD and Intel in the PC market, with both companies acknowledging the significance of GPUs, which are integral for handling complex computations.
• 09:00 - 12:30: NVIDIA's AI Focus and CUDA This chapter delves into NVIDIA's strategic emphasis on artificial intelligence (AI) and the role of their parallel computing platform and application programming interface (API), CUDA. It also contrasts with the integrated and competitive landscapes of AMD and Intel, focusing on AMD's acquisition of ATI to bolster their position in the graphics segment and the implications this had for their market strength. The context highlights NVIDIA's distinct technological focus and strategic foresight in advancing GPU capabilities tailored for AI workloads.
• 12:30 - 16:30: AMD's Financial Challenges and Expansion Efforts The chapter discusses AMD's financial difficulties and its strategic decisions to expand in the competitive tech market. The company explored modular processor designs combining CPU and GPU capabilities to tap into future computing needs. Despite this, skepticism existed due to AMD's significant financial constraints, primarily its debt obligations, which were substantial by mid-2007.
• 16:30 - 21:00: AMD's Attempts to Catch Up in AI The chapter discusses AMD's financial struggles and attempts to catch up in the AI industry. Despite introducing new graphics products from ATI, AMD faced significant losses in 2007 and 2008, attributed to a price war with Intel which resulted in reduced average selling prices, even though they were shipping more units. The acquisition of ATI did not yield the expected benefits.

AMD's $243 Billion AI Disaster...What Happened? Transcription

• 00:00 - 00:30 Thanks the AI boom, tech giants are pouring  billions into chips like GPUs. This propelled   NVIDIA to the top, briefly ranking as  the most valuable company in the world.  But, what about AMD? Like NVIDIA, they’ve been designing   semiconductors, more specifically GPUs for some  time. They should be dominating, but aren’t.  In fact, their shares have recently plunged. From  early 2024 to early 2025, their shares dropped by
• 00:30 - 01:00 half, erasing almost $200 billion in market cap. What’s going on?  How is a GPU company failing during the AI hype? What is happening to AMD? To answer this, we need to take a look back. In the 2000s, AMD was in a fierce battle with   Intel. Both these companies produced  CPUs, central processing units,
• 01:00 - 01:30 and had gone back and forth as the market  leader. But now, Intel was pulling ahead.  They had their factories to produce CPUs  at a lower cost, and at higher volume   than AMD, who relied on GlobalFoundries. They had secured great partnerships with   giant customers like Dell and HP. But there  was something much more sinister going on.  Reportedly, Intel paid their customers billions  behind the scenes to not use AMD. Intel at the   time was making much, much more money than AMD,  and had the cash to try and crush AMD completely.
• 01:30 - 02:00 Lawsuits from AMD aimed at Intel ensued,  arguing anti-competitive practices.  The Japan Fair Trade Commission had found that  Intel was, indeed, paying such rebates to five   major Japanese computer makers (presumably  Sony, Toshiba, NEC, Hitachi, and Fujitsu,   though the companies are unnamed in the public  version of the JFTC order) and that the rebates
• 02:00 - 02:30 violated Japanese competition law. In its suit AMD  alleges that Intel has been paying manufacturers   so-called first-dollar rebates, meaning that  at the end of the quarter, if the customer has   achieved the level of exclusivity Intel seeks,  it will get a retroactive discount on every Intel   processor it purchased that quarter. Yet there was another major issue.  Tension Both companies realized a growing part   of the PC market was in another technology:  GPUs, which was important for complicated
• 02:30 - 03:00 computing involving graphics: like video games. Though not comparable to standalone GPUs, Intel   was using integrated graphics with their CPUs,  so they had something while AMD only made CPUs.  AMD needed to do something. They were  barely profitable, and losing ground.  So, AMD made a gamble. They bought the Canadian   GPU maker “ATI”, for $5.4 billion. In a press release AMD "In this increasingly   diverse x86 computing environment, simply adding  more CPU cores to a baseline architecture will not
• 03:00 - 03:30 be enough. Modular processor designs  leveraging both CPU and GPU compute   capabilities will be essential in meeting the  requirements of computing in 2008 and beyond.”  This would give them an “in”  into this growing market.  But most people were sceptical. This was money AMD didn’t have,   and as a result, almost all of that went to debt. In mid 2007 their total debt was about as much as
• 03:30 - 04:00 their total revenue that year. But things didn’t improve.  They made a $2.8 billion loss in 2007, as their  “computing solutions” business fell 12%, even with   new graphics products from ATI. Reportedly,  reasons included a price war with Intel.  “Even though AMD was shipping more units,  its average selling prices were falling.”  Then in 2008, they lost another $2.3 billion. Even then, the ATI acquisition didn’t work out as
• 04:00 - 04:30 intended. They were having trouble merging  the companies, and the lawsuits against Intel   were a massive distraction. It also  appeared that they overpaid for ATI,   which was having its own problems, including delay  of graphics products and their most important   customer: Motorola, giving them less business. By 2010, AMD dropped the ATI brand altogether.  Things were bad. Even though they had technically won   in court against Intel, with a $1.25 billion  settlement, they still were in a bad spot.
• 04:30 - 05:00 But luckily, someone came to the rescue. Enter Lisa Su. Su was an engineer, with deep  academic and field experience.
• 05:00 - 05:30 She had a vast amount of experience with  semiconductors and chips, coming from Freescale   Semiconductor, IBM, and Texas Instruments. Su was made the senior vice president and general   manager in 2012, and soon was appointed to CEO,  largely because everyone above her dropped out.  She had a lot of experience,  but a massive challenge.  She had to get AMD back on track. Luckily, she had an answer.
• 05:30 - 06:00 In truth, even before she was made CEO, she  was already working to save the company.  You see, AMD was stagnant, yet also  extremely narrow: The PC market.  But, that market wasn’t growing. In 2011, PC shipments only grew by   0.5%, then they began to decrease. In  2013, shipments declined by over 10%!  AMD was pigeonholed into a  market that was going down.
• 06:00 - 06:30 But, it didn’t have to be this way. CPUs and more recently GPUs, were their   business and both of those categories,  and for that matter, semiconductors,   have a huge spread of industries. This was Lisa’s plan. She began meeting with  leaders at Microsoft and Sony.  Her focus was diversifying the company.  Expanding from the PC market into growing   industries: Namely: consoles. Through her relationships,   she secured AMD as the backbone of the new  generation of consoles: The Xbox One, and the
• 06:30 - 07:00 PlayStation 4. Each would have custom AMD chips. This was the 8th console generation, and the   industry was booming as it had been for a while. To date, the Xbox One has shipped about 58 million   units. Impressive, except, when compared to the  PlayStation 4, which shipped over 117 million.  All of them are powered by AMD chips. In fact, with how many PS4s have been sold,   you could argue Sony saved AMD from Bankruptcy. When Lisa Su joined in 2012, just 10% of AMD’s
• 07:00 - 07:30 sales came from Non-PC markets, but  in just 3 years, that climbed to 40%.  This was a monumental success. AMD was  making more profit, and was much more stable.  But in reality, this was just  the beginning of AMD’s comeback.  With some momentum behind them,  it was time for AMD to take back   ground in their home market: Computers CPUs. In 2017, AMD launched their next biggest brand:
• 07:30 - 08:00 Ryzen. CPUs with high performance, high thread  counts, and unbeatable value for the dollar.  In just a few months, AMD’s CPU market  share surged, and Intel’s began to decline.  AMD had tighter profit margins, but it was a  price they were willing to pay to beat Intel,   who, at this point, were starting to panic. They were having huge delays with their new   10nm chips, which were supposed to launch  in 2016, but pushed as far back as 2019.
• 08:00 - 08:30 To add salt to Intel’s wound, while they  were struggling to produce 10nm chips,   AMD had introduced 7nm chips, though  I will note that they each measure   their process sizes slightly differently. But a key reason for this was manufacturing.  Their inhouse chip production had provided better  products and better profits in the past, but now,   it was hurting them. AMD however, had moved from  GlobalFoundries to TSMC in 2018, who were far
• 08:30 - 09:00 further ahead in producing 7nm chips. AMD, for the first time in decades,   was beating Intel. Lisa Su had done it,   and there was more than just Ryzen. AMD had also cemented itself as a   cheaper yet still great alternative to NVIDIA  in GPUs, with extremely good value for money.  She had also pulled back AMD from  fields that were just costing money.
• 09:00 - 09:30 In November 2012, AMD’s share price  was as low as $2, but by that time   of year in 2020, it had climbed to $92. Lisa Su had spent years directing AMD’s   focus and investments into the future,  which were now paying off, big time.
• 09:30 - 10:00 Everything was going great… or was it? Lisa Su had to make these early bets,   and they worked… at least for the  medium term, but not the long term.
• 10:00 - 10:30 As it turns out, while AMD was  making a CPU and PC comeback,   Nvidia was putting their focus  somewhere very different… Lisa Su is one of the most powerful women  in tech—but with a simple Google search,   you can find her home address, personal email,  and phone number. And—it’s not just Lisa.
• 10:30 - 11:00 You can look up almost anyone’s  personal info online, including   yours, and that’s why I’m excited  about today’s sponsor, Incogni. Incogni is a personal data removal service that  automatically reaches out to data brokers—those   sketchy companies that collect and sell your  personal info—and gets your data taken down. After someone used my credit card  to run a bunch of Facebook ads,   I went down a rabbit hole  trying to clean things up. I found dozens of sketchy people search  sites that had my name, old addresses,
• 11:00 - 11:30 even emails I forgot existed—but  getting off those lists was a   nightmare. Every site had a different  process if they let you opt out at all. Incogni makes the whole process  easy—handling the requests,   the follow-ups, and even pushing back when data  brokers resist. And if your data shows up later,   they’ll remove it again, which is what  makes the annual plan so worth it.  These brokers fuel everything from robocalls  to phishing scams to identity theft. It’s one
• 11:30 - 12:00 of those things you intend to take care of  yourself... but realistically, never will.  If you want to take back your privacy,  check out Incogni at the link below. It   supports the channel, and if you use code  LOGICALLYANSWERED, you get a 60% discount. Thank you to Incogni for supporting our videos. While NVIDIA was involved in gaming and  general computing. Unlike Intel and AMD,   who were in a great battle over PCs, NVIDIA  were deeply focused elsewhere: Research. And,
• 12:00 - 12:30 it wasn’t for the short term, or even the  medium term. It was for the distant future.  You see, they have been working towards one goal  for a long time. For decades even: Generative   learning. And the first step for this was as far  back as 2006, even if NVIDIA didn’t realize it. GPUs are extraordinarily powerful,  but highly complex. They can perform   an unbelievable amount of math, but the  problem was using them. Or rather, building
• 12:30 - 13:00 software that could take advantage of that power. So, NVIDIA released CUDA (Compute Unified Device   Architecture). Essentially, a powerful API for  their GPUs. CUDA allowed users to build code in   languages like C++ or Python to leverage NVIDIA’s  GPUs, but this was for non-graphics tasks.  “CUDA made it vastly more accessible for  engineers, researchers, and scientists
• 13:00 - 13:30 to leverage GPU acceleration.” Because CUDA was made so early,   and had consistent, unwavering support in  the field of research, it became the default.  NVIDIA became the preferred GPU provider,  not just because their GPUs were good,   but because of the ecosystem around them. So how does this play into AI?  Well, let me answer that with one crucial  example. The Annual ImageNet Challenge.  This was an AI competition, where researchers  tried to build the best image recognition model,
• 13:30 - 14:00 by viewing millions of images  across thousands of categories.  Most teams entered with large amounts of CPUs  to power their neural networks, except one…  In 2012, Alex Krizhevsky, a PhD student at  the University of Toronto, entered. But his   model was a little bit different. Alex was using GPUs. Or rather,   just two NVIDIA GTX 580s. And he won by a   landslide. (Alex’s team “SuperVision)” This was one of the most pivotal moments
• 14:00 - 14:30 for AI, and showed the world that the  future was not through CPUs, but GPUs.  But this was only possible, thanks to CUDA. The model was later refined and named “AlexNet”,   and Alex’s later research start-up  was eventually acquired by Google,   but as for NVIDIA, they began to pour more  and more resources into preparing for AI.  While AMD was winning the war of CPUs, NVIDIA  were still building AI architecture. Still
• 14:30 - 15:00 banking on this future idea. In Nvidia’s 2016 report,   they stated “Deep-learning breakthroughs  have sparked the Al revolution. Progress is   exponential. Adoption is exponential.  The impact to the tech industry and  society will also be exponential.”.  (Shareholder Report 2016).  Keep in mind, this is many years  before ChatGPT, and OpenAI was   only founded a matter of months ago. At this time, NVIDIA was already releasing   AI and deep learning designed GPUs, like the  Tesla P100 series with Pascal architecture.
• 15:00 - 15:30 Then the following year, NVIDIA released the  Tesla V100, along with Volta Architecture,   which could perform calculations 12x  faster than Pascal. (Source: page 21).  These advancements in machine  learning continued year after year.  AI wasn’t even in the mind of  the public yet, or anyone really. This brings us back to AMD and their  investments, which were going elsewhere.
• 15:30 - 16:00 At the start of the AI boom, AMD did see a massive  jump. Their revenue skyrocketed, from $9.7 billion   in 2020, to $23 billion in 2022. But then, things get interesting.  Their revenue dropped the following year. And  their profit shows something even more alarming.  2021 was their most profitable  year in history, at $3.7 billion,
• 16:00 - 16:30 but things take a nosedive, back down  to $1.2 billion the following year.  I think this is when researchers  came to the big realization.  AMD wasn’t the answer for AI. NVIDIA  was the far, far better choice.  Their GPUs were good, but the software support  just wasn’t there. Their Radeon GPUs were used in   some machine learning, but had nothing like CUDA. By the time AI began to show its real potential,
• 16:30 - 17:00 researchers began investing  into GPUs, and lots of them.  When looking at AMD, Intel, and NVIDIA, one  of them was the clear choice. Not only for the   physical chips, but also the architecture  and many, many years of prior research.  If something has become the standard for progress,  it’s very hard to disrupt the status quo.  Things continued to drop for AMD, with profits  falling again to only $600 million in 2023.  To be fair: They had invested in cloud computing  and datacentres, but not so much machine learning.
• 17:00 - 17:30 Lisa Su, after saving AMD, now had  an even bigger challenge ahead.  AMD needed to catch up. But how? AMD went into AI overdrive. In early 2022, AMD acquired Xilinx.  Lisa Su said that “AMD will be able  to increase its breadth in key markets   like data centres where Xilinx has a  strong network and Al presence” (source)  But this wasn’t just any acquisition. Xilinx cost  AMD $50 billion. And… AMD didn’t have $50 billion
• 17:30 - 18:00 lying around. Was this going to be a repeat of the  ATI situation? Was their debt going to skyrocket?  Instead, they performed a “stock transaction”,  where Xilinx shareholders received millions   of AMD shares. But this, naturally,  wasn’t good for their current investors.  This was also right before  AMD’s profits began to plummet.
• 18:00 - 18:30 Investor confidence sank, and their share price,  which was as high as $143 at the end of 2021,   tumbled all the way to $84 by August 2022. But AMD marched on.  On October 10th, AMD announced it had  acquired Nod.AI, an open source AI provider. .  AMD in a press release stated the acquisition  would “significantly enhance AMD’s ability to   provide customers with “software that  allows them to easily deploy highly   performant AI models tuned for AMD hardware.” We don’t know how much they paid for Nod.ai,
• 18:30 - 19:00 but the reason was clear: AMD  was trying to catch up to CUDA.  Then, on November 15th, they announced  their next big move: The MI300 series.  Data centre accelerators, which used AMD’s new  architecture; CDNA 3, which had just launched.  GPUs specifically designed for AI workloads. These were their chips to rival NVIDIA,   and already, it seemed like  things were turning around.  They secured customers like  Microsoft, Dell, and Meta,
• 19:00 - 19:30 who would use the MI300 series for AI training. Su commented that “We are seeing very strong   demand for our new Instinct MI300 GPUs, which  are the highest-performance accelerators in   the world for generative AI. We are also  building significant momentum for our data   center AI solutions with the largest cloud  companies, the industry’s top server providers,   and the most innovative AI startups” AMD was pouring billions and billions   into AI. New chips, new architecture, and  all kinds of acquisitions and investments.
• 19:30 - 20:00 A lot of big moves, but what  actually came out of it? Unfortunately, things didn’t go as AMD hoped. The MI300 series was expected to generate   $8 billion, but pretty soon,  that was dropped to $5 billion.  It still wasn’t quite as good as NVIDIA. In late 2024, AMD hosted a massive event   called “Advancing AI 2024” but the reception  was mixed, and analysts called it “largely
• 20:00 - 20:30 uneventful”. “We see additional downsides  as we now believe its AI GPU roadmap is less   competitive than we previously thought,” “Hence, we believe AMD wouldn’t be able   to penetrate the AI GPU market as  much as we had earlier anticipated.”  AMD is desperately trying to catch NVIDIA.  But also, their investors expectations.  Analyst Rick Schafer said "We applaud the  progress AMD has made with MI Instinct,   growing a roughly $4 billion AI franchise  from nothing in just 12 months," Schafer
• 20:30 - 21:00 said. "Unfortunately investor expectations  have remained persistently out of reach.” AMD had a good focus for the short term,   but it wasn’t amazing for the long  term. And it’s hard to blame them.  They spent so long getting out of the red, by the  time AI actually hit, they had to play catch up.  In the middle of their past challenges,  they lacked profit margins or the   capital to pour into early AI infrastructure. NVIDIA however, were preparing the technology
• 21:00 - 21:30 long before the market was ready for it. Can AMD possibly catch them? Well,   they’re certainly trying to. But I’m not so sure. Lisa Su is a great leader. But it seems she’s in   the same situation now as she was over 10 years  ago: Trying to improve a difficult situation.  Ironically, Intel is in an even worse  position than AMD having bet $100 billion   on making their own chips which has largely  backfired. Check out this video to learn more.